Download Time to model all life on Earth - Department of Mathematics and

COMMENT
ECOLOGY Zoological
travelogue tracks rare
species worldwide p.300
WOMEN Calls to root out sexism
in journals, conferences and
experiments p.305
OBITUARY Rita Levi-Montalcini,
nerve growth factor pioneer
and science advocate p.306
YI LU/VIEWSTOCK/CORBIS
COMMUNICATION Sally Rockey
reflects on two years of
blogging at the NIH p.298
A hyena surveys a flock of flamingos in South Africa.
Time to model
all life on Earth
To help transform our understanding of the biosphere, ecologists — like climate
scientists — should simulate whole ecosystems, argue Drew Purves and colleagues.
N
o report from the Intergovernmental
Panel on Climate Change would fail
to mention global climate models.
Yet the international bodies that are charged
with addressing global challenges in conservation — including the Intergovernmental
Platform on Biodiversity and Ecosystem
Services, which holds its first plenary meeting next week in Bonn, Germany — cannot
refer to analogous models of the world’s ecosystems. Why? Because ecologists have not
yet built them.
General circulation models, which simulate the physics and chemistry of Earth’s land,
ocean and atmosphere, embody scientists’
best understanding of how the climate system
works and are crucial to making predictions
and shaping policies. We think that analogous
general ecosystem models (GEMs) could radically improve understanding of the biosphere
and inform policy decisions about biodiversity and conservation. Currently, decisions
in conservation are based on disparate correlational studies, such as those showing that
the diversity of bird species tends to decline in
deforested landscapes. GEMs could provide a
way to base conservation policy on an understanding of how ecosystems actually work.
Such models could capture the broad-scale
structure and function of any ecosystem in
the world by simulating processes — including feeding, reproduction and death — that
drive the distribution and abundance of
organisms within that ecosystem. Ecologists
could apply a GEM to African savannas, for
instance, to model the total biomass of all the
plants, the grazers that feed on the plants, the
carnivores that feed on the grazers and so on.
Over time, the flows of energy and nutrients
could be mapped between them. All of the
organisms would be grouped not by species,
but according to a few key traits such as
1 7 JA N UA RY 2 0 1 3 | VO L 4 9 3 | N AT U R E | 2 9 5
© 2013 Macmillan Publishers Limited. All rights reserved
whether they are plants, birds or mammals,
cold blooded or warm blooded, diurnal or
nocturnal. By encoding processes such as
migration and predation into simple mathematical and computational forms, ecologists
could model what happens to the various
groups over time.
Metrics such as the diversity of animal
types inhabiting the grasslands could be used
to assess the savannas’ health, stability and
resilience, and to analyse the fate of particular groups of organisms such as top predators.
Ecologists could explore how these attributes
might change in response to, say, climate
change, the introduction of invasive species or
poaching. And, because the rules of play are
likely to be broadly similar no matter what the
ecosystem, the GEM could equally be applied
to forests, lakes or the remotest parts of the
ocean, providing a common framework for
understanding and managing disparate ecosystems on local and global scales.
There are huge challenges to building
GEMs — not least, obtaining the appropriate types of data to validate the models’
predictions. But the difficulties are not insurmountable. Theories abound for describing
the processes that drive ecosystems, many of
which are backed up by data.
BUILDING A PROTOTYPE
Over the past two years, we at Microsoft
Research and at the United Nations Environment Programme World Conservation
Monitoring Centre, both in Cambridge, UK,
have built a prototype GEM for terrestrial and
marine ecosystems. Called the Madingley
model, it uses real data on carbon flows as a
starting point. We have hit all sorts of computational and technical hurdles, and are
expecting more as we develop the model. Yet
the project demonstrates that building GEMs
is possible. From the relationship between the
mass of individual organisms and how long
they live, or the effects of human perturbations such as hunting, to the distribution of
ecological complexity is so obvious in nature.
Ignoring the myriad shapes and colours of
different bird species, for example, seems
instinctively wrong. Instead, ecologists have
tended to stress the importance of species
identification as well as a vast number of
ecological processes, from individual adaptation to the social dynamics of groups. Many
in the field also emphasize that findings in
one ecosystem do not generalize to others,
and that randomness and history could be as
important in affecting some particular measurement as any deterministic rules3. But comprehensive species-specific data will always be
in short supply (at least 80% of the millions of
species on Earth are undescribed4), so a better understanding of ecosystems demands a
broad-brush approach.
Building a GEM will require different
types of data — to help define the ecological
rules at play, to provide reasonable starting
conditions for the simulation (such as a realistic ratio of herbivores to top predators) and
to evaluate the model’s predictions. For these
three goals, enough data are available to get
started, although information on ecological
processes far outweighs the rest.
Metabolic rates, for instance, have been
measured in hundreds of animals in the
lab, and researchers in the field have documented life­spans, growth rates and reproductive success for thousands (in some cases,
millions) of birds, mammals, plants and bacteria. Ecologists have also mathematically
determined numerous ‘rules of existence’
for some organisms, such as that an animal’s
metabolic rate is proportional to its mass
raised to a power of around 0.70 (ref. 5).
Modelling every organism in an ecosystem such
as a tropical rainforest would be impossible.
biomass across Earth (see ‘Model life’), the
model’s outputs are broadly consistent with
current understanding of ecosystems.
Modelling ecological phenomena at various scales is not new. Conservationists frequently use models to predict how much
habitat fragmentation an endangered species
can tolerate. But few ecologists have tried
to build models of how general ecosystem
properties emerge from the interactions of
individuals. One attempt — known as Ecopath with Ecosim1 — is being developed at
the University of British Columbia’s Fisheries Centre to address management and ecological questions. This combines modelling
at the phytoplankton scale with that at the
level of fisheries and marine mammals.
Another, called Atlantis2, developed by Australia’s Commonwealth Scientific and Industrial Research Organisation, incorporates
biophysical, economic and social factors
to provide an integrated tool for modelling
marine ecosystems. But these consider fewer
processes and organisms than would a GEM.
Throughout the history of ecology, most
researchers have resisted abstraction because
MODELLING BEHAVIOUR
Obviously, modelling every organism within
an ecosystem is impossible. (We estimate
that it would take a standard laptop computer around 47 billion years to model for
100 years every multicellular animal within
just one of the 1-degree grid cells covering
MODEL LIFE
Variation in biomass across the world simulated by the Madingley model for terrestrial and marine ecosystems. Fundamental ecological processes, encoded into simple
computational forms, determine the abundance and body mass of organisms (grouped into cohorts for simplicity) and so indicate the state of ecosystems.
Ecosystem
state
Abundance
Herbivore
cohort
Carnivore
cohort
Omnivore
cohort
Body mass
High biomass
Ecological
processes
Low biomass
Reproduction
Eating
2 9 6 | N AT U R E | VO L 4 9 3 | 1 7 JA N UA RY 2 0 1 3
© 2013 Macmillan Publishers Limited. All rights reserved
Metabolism
Dispersal
Mortality
Other
SIMON LONG/FLICKR/GETTY
COMMENT
ZEBRA: MANOJ SHAH/GETTY;
WHALE SHARK: ANDY ROUSE/NATUREPL.COM
COMMENT
From hunting zebra to filter feeding, the process of predation in all ecosystems plays by similar rules.
Earth.) Yet certain computational techniques
have been developed, mainly in marine ecology, that could allow researchers to model
entire ecosystems using rules about the
behaviour of individuals.
One approach is to model collections of
organisms or ‘cohorts’. The idea is that within
a cohort, individuals are similar enough to
be considered identical. For a shoal of small
herbivorous fish meeting a cloud of plankton, say, ecologists could calculate the feeding rate for an exemplar fish and then apply
that rate to the whole shoal. (In the simulation just described, we found that grouping
organisms into cohorts on the basis of body
size, functional group such as omnivore or
carnivore and a few other traits reduced the
computation time to 10 hours).
The biggest stumbling block to constructing GEMs (after convincing ecologists that
they can and should be built!) is obtaining
the data to parameterize and validate them.
Records of what species of plants and
animals live in the world’s forests, grasslands and oceans are often available to
some extent, but far fewer data exist on the
abundance of those species. And almost
no data have been collected on the properties of whole ecosystems, such as on the
distribution of body sizes from plankton to
whales. Marine trawl surveys carried out
for research or to assess fish stocks probably
come closest to providing this type of information, although even these are restricted as
to what size range of organisms they survey.
A new programme of data gathering is easy
to envisage. Using automated cameras and
image recognition, it should be possible to
sample thousands of animals and determine
their approximate size and what broad group
they belong to: reptile or mammal, flying or
non-flying. Motion-activated cameras used
by conservationists and wildlife enthusiasts
already produce tens of thousands of images
of fish, birds and mammals every day. And
stored away in numerous research institutions are vast samples of insects collected in
traps that suck them out of the air, and data
from continuous plankton recorders towed
beneath ships for millions of kilometres.
Naturally, a major new data-gathering
programme would be costly. But globally,
governments already spend billions of dollars on satellite observations of vegetation
and habitat distribution, fisheries surveys,
forest inventories and species surveys.
Diverting a small fraction of these funds to
gathering the data needed to develop and
evaluate GEMs could pay dividends. A first
step would be for governments around the
world to support programmes similar to the
National Ecological Observatory Network
— an international cooperation funded by
the US National Sci“Ecological
ence Foundation to
systems do
manage large-scale
not have the
collection of ecological and climate data.
equivalent of
To reduce costs and
the precise
to harness the power
laws used
of citizen science, data
by climate
collection could even
scientists.”
be crowd-sourced.
Rapidly growing websites such as iNaturalist and eBird (on which users can share their
observations of wildlife) currently focus on
traditional species identification. Such sites
could potentially collate an extraordinary
amount of information on functional groups
of organisms and traits such as body size.
TRUSTED ADVICE
Constructing realistic GEMs is one thing.
The real challenge is to produce models from
which the predictions are trustworthy enough
to guide the decisions of conservationists and
policy-makers. Recent progress in computational statistical methods offers a way for
ecologists to formally build trustworthiness
into models. For instance, tools are available to quantify the uncertainty associated
with models’ predictions. A healthy crop of
alternative, competing GEMs will be crucial,
together with mechanisms that enable their
fair assessment. In blind-testing, for example,
different models could be used to predict an
ecosystem property that has been measured
but not reported, allowing the models to be
ranked in terms of how well they do.
We are not proposing that GEM predictions (which will always be simplistic) provide the only guide to conservation policy
and the management of ecosystems. But
coupled with models from other fields, such
as economics and epidemiology, they could
offer a means of managing human actions
and the biosphere in an integrated, consistent
and evidence-based way. Far from eclipsing
traditional ecological research, GEMs would
draw on it and give such work more focus.
Using GEMs, ecologists could identify processes that are poorly understood yet crucial
to ecosystem structure and function, rather
than delve deeper into well-studied areas.
Ecological systems do not have the equivalent of the precise laws used by climate
scientists. This is a significant challenge to
building GEMs, along with the complexity
of nature, the small number of GEM-like
models under development and the paucity
of data with which to constrain them. But
just by attempting to build general models,
ecologists will find out what they need to
know to truly understand ecosystems. ■
Drew Purves is head of the Computational
Ecology and Environmental Science Group
at Microsoft Research in Cambridge, UK.
Jorn Scharlemann, Mike Harfoot, Tim
Newbold, Derek P. Tittensor, Jon Hutton,
Stephen Emmott.
e-mail: [email protected]
1. Christensen, V. & Walters, C. J. Eco. Mod. 172,
109–139 (2004).
2. Fulton, E. A. et al. Fish Fish. 12, 171–188 (2011).
3. Boyd, I. L. Science 337, 306–307 (2012).
4. Mora, C., Tittensor, D. P., Adl, S., Simpson, G. B. &
Worm, B. PLoS Biol. 9, e1001127 (2011).
5. Brown, J. H., Gillooly, J. F., Allen, A. P., Savage, V.
M. & West, G. B. Ecology 85, 1771–1789 (2004).
Full author affiliations accompany this article online
at go.nature.com/ob2a2p.
1 7 JA N UA RY 2 0 1 3 | VO L 4 9 3 | N AT U R E | 2 9 7
© 2013 Macmillan Publishers Limited. All rights reserved